The sparse code multiple access (SCMA) systems rely on multiuser detection schemes to achieve high spectral efficiency and massive connectivity. The existing multiuser detection schemes employ a message passing algorithm (MPA) to achieve the goal by utilising the sparse structure of the SCMA codebook. However, it remains a challenging task to obtain a good trade-off between the BER performance and computational complexity due to the iterative nature and the associated search space in the MPA detection. In this Letter, a novel MPA detection scheme based on selective message passing is proposed. By dynamically eliminating the branches with high reliability from the message passing process, the proposed scheme significantly reduces the computational complexity with negligible BER performance compared with the conventional MPA detection. Simulation results show that the proposed scheme is able to achieve a good trade-off between the BER performance and computational complexity, especially at moderate and high signal–to-noise ratio conditions.